[gmx-users] How can I separates only the statistically independent configutation?

Eudes Fileti fileti at iq.usp.br
Sat Aug 12 13:36:02 CEST 2006

Dear Gmx users

In my phD I applied  the QM/MM sequential methodology.
Differently of the QM/MM conventional methodology,
here we performed the quantum calculations after the
simulation and not simultaneously (see some refs:
Phys. Rev. E, 67, 61504, 2003; J. Chem. Phys.,117, 1692, 2002).
We carry quantum calculation only over
the configurations of the simulation that are structurally
uncorrelated, are they that have statistical relevance.
With Monte Carlo, we use the self-correlation function
of the energy. This function says us what the interval of MC steps
we must collect a configuration so that this does not have no correlation
with the previous one. In the case of Molecular Dynamics this is
make by van Hove function. In this case the function of van hove
say us what time interval I must save a form structure that this if
does not remember the previous one. Thus, we will only take to the
quantum calculation the configurations with relevance statistics.

Dr van Maaren have suggested me apply something
like the block averaging method to the msd to get some kind of feeling
how large the interval should be? I did this using:

g_analyze -f potential.xvg -ee

When I following its suggestion to use the process
of block-averaging to uncorrelate my structures
I realized that I could use the autocorrelation function
of the potential energy. For this I made

g_energy -f ener -s term -o potential

After that I made

g_analyze -potential -ac acorr

However I have not obtained the standard
like the figure 2 of Mol.Phys, 1986,57,89
(exponetially achieving the zero of C(i) )
but, on the contrary of this, I got an oscillation,
giving positive and negative values for C(i).

I assume that this behavior is not correct
and I cant to perceive my error. (If interesting I can
sent to you, not to list, the chain of potentil energy
with 40000 time steps)

As a last attempt, I tried to found the van Hove function at
Gromacs but some posts at maling list indicates me that
this function is not implemented into Gromacs.

SUMMARIZING: I have one trajectory of 800ps. I need extract
this trajectory uncorrelated (statistically independent)
configurations. So I need know how many ps separates two consecutive
configuratios. Which procedure I can use for this analysis?

                                      Eudes Eterno Fileti
          Centro de Ciência Naturais e Humanas
                        Universidade Federal do ABC
                                  Rua Santa Adélia, 166
                                            CEP 09210-170
                                               skype: eefileti
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20060812/671b9369/attachment.html>

More information about the gromacs.org_gmx-users mailing list